Net Promoter Score benchmarks: What good looks like by industry

A target built from your own baseline is what you can defend when the board asks whether 28 is good. Your Net Promoter Score (NPS) came in at 28 this quarter, and three benchmark reports place your industry in three different positions. A second question follows: why phone wait times rose that quarter.

The board judges the AI agent pilot on the voice channel by the enterprise score, with no separate score for the calls it handled. Published Net Promoter Score benchmarks cannot settle whether the score is good or explain the longer wait times, because each report measures something different and rarely says what.

What NPS is and how it is calculated

NPS is a loyalty metric. To calculate it, ask customers how likely they are to recommend a company on a 0-10 scale and subtract the percentage of detractors (0-6) from the percentage of promoters (9-10). Passives, who answer 7 or 8, remain in the sample but do not enter the subtraction, so the result runs from -100 to +100.

The question asks about the company as a whole. It does not ask which interaction the customer had in mind, whether they called last week, or whether they have ever contacted service at all. The question's broad scope helps explain why published figures for one industry rarely agree. Safe comparisons require study parameters that identify what each score measures; matching those parameters prevents an invalid broad comparison.

The three NPS types every benchmark hides

Before you can read any benchmark fairly, establish which NPS it reports. Cross-source divergence within one industry and one year almost always traces back to measurement design, and the first place that design shows up is the NPS type the survey captured. A post-call survey and a brand survey both produce a number between -100 and +100, but they measure different things and cannot be pooled or compared without distortion.

A useful taxonomy separates three types:

• Relationship NPS: the score at brand level against competitors, drawn from customers rating the company as a whole rather than a specific interaction.

• Episode NPS: the score for a defined experience such as a call, a claim, or an onboarding, which identifies the moments that create promoters or detractors.

• Channel NPS: the score for a delivery channel, which compares in-person against online or voice against digital performance.

Round-ups that pool these under one industry label compare scores that never measured the same thing. When you read a benchmark, find the type first, then decide whether it belongs next to your number.

Where industries actually land on relationship NPS

A relationship NPS below 0 signals more detractors than promoters and points to broader problems with the customer relationship. A score between 0 and 30 is a good range, though it leaves real room for improvement. Above 30, a company is doing great, with far more satisfied customers than unhappy ones. Above 70, a company sits in world-class territory, generating strong positive word-of-mouth. With that scale in mind, here's where the phone-heavy service industries below actually land.

The 2024 Q3–Q4 U.S. Consumer Benchmark Study is the most recent credible NPS source for industry benchmarks. The study surveyed 10,000 consumers in 354 companies and 22 industries, measuring brand-level loyalty in general, not post-call sentiment. Use the industry scores as a sanity check for your own score, never as a target, and treat a newer figure elsewhere as possibly measuring a different NPS type until its method says otherwise.

These industries reflect the categories most reliant on the phone channel and enterprise contact centers, where call volume, resolution rates, and routing performance most directly shape relationship NPS.

To evaluate the phone channel, collect a channel-level or relationship NPS from your own survey data using a consistent NPS type and population.

The same Qualtrics study places utilities and TV/internet service providers at the bottom of the range shown here. Inbound service contact can be routine in those categories: a billing dispute or an outage. Retail customers may need to call service less often. Industry rank therefore cannot set a phone-channel target; the contact center needs a matched internal baseline before leaders judge performance.

Design choices that create cross-source divergence

Even after you pin down the NPS type, the same industry can show wildly different scores across publishers because of the choices made when the survey was designed and the data was pooled.

Four design choices account for most of that divergence, and each one is easy to check before an external number goes next to yours in a board deck. Screening a benchmark against them takes minutes and prevents the wrong comparison from anchoring a target.

• NPS type blended: Publishers average relationship and episode scores from different surveys into one industry figure, but scores measuring different parts of the customer relationship should remain separate.

• Population mix: Business-to-business (B2B) and business-to-consumer (B2C) respondents, or enterprise and small-business customers, sit inside one published number, and an enterprise insurer's book of business looks nothing like a respondent pool consisting mostly of small-business policyholders.

• Self-reported versus independently measured: A round-up may pool scores that companies submitted about themselves with scores that researchers fielded independently to customers, and self-submitted figures carry an obvious selection problem.

• Sample too small: Small channel samples may not distinguish modest NPS changes from sampling variation under conventional confidence and power standards.

Run your own program through the same four checks before an external number goes next to it in a board deck. NPS earns its place as one input in a voice of customer program alongside transactional measures and operational data, and the operational data is the part a CX leader can act on this quarter.

The contact-center metrics that move NPS

Relationship NPS lags what happens on the phone, because a customer answering a brand survey in March is drawing on the claim they filed in November and the two calls it took to close it. Enterprise contact centers often find that detractors cluster around unresolved and repeated contacts, so the score a board sees is a delayed readout of resolution and routing performance.

These operational metrics translate directly into the promoter and detractor counts that eventually reach the brand survey, and each one is measurable in the current period:

• First call resolution (FCR): The share of contacts closed on the first attempt. Callers who do not get a resolution have to come back, and repeat callers are where detractors cluster.

• Average handle time (AHT): Total handling time per interaction, including the conversation, hold periods, and after-call work. A short call that ends in a transfer lengthens the customer's total time to resolution, so read AHT alongside FCR rather than against it.

• Containment rate: The share of contacts resolved without a human agent. Containment removes routine volume from queues so the remaining human-handled calls wait less, but it helps the score only when the AI agent actually resolves contained calls.

• Routing accuracy: The share of contacts that reach the right skill on the first attempt. Every misroute adds more hold time and forces the customer to explain the problem again after a transfer.

On the phone channel, two mechanisms decide where a call lands on those metrics. Intent recognition accuracy determines whether a caller who says "I need to change the payout account on my claim" reaches the claims skill team on the first attempt or gets parked in a general queue. Escalation logic determines whether an AI-handled call that needs a human ends as a warm handover with the caller's context intact, or as a hang-up followed by a second call the next morning.

Post-call customer satisfaction score (CSAT) measures one contact, and improving CSAT on a single call feeds the relationship score over time. Wait time drives the call abandonment rate, and callers who hang up often don't enter the standard post-call survey flow, so operational metrics can move within a month while the brand score arrives quarters later.

How to set NPS targets around an AI agent deployment

A CFO's first question is why 35 and not 30, and what it costs to get there. Only a target you build from your own baseline and a costed lift answers it when you break out results by who handled the call. A five-step target-setting sequence turns external industry benchmarks into a defensible internal number.

1. Establish the baseline

Establish your current score for one NPS type and one population, and forecast where the trend puts it next year if nothing changes. The baseline anchors every later number, so keep the population definition and survey method constant across the periods you plan to compare. A moving baseline invalidates the entire sequence.

2. Define the desired lift

State the increase you want above that forecast, in points, for a defined period. The lift is not the target score itself; it is the movement above the trend line, which separates gains from what would have happened anyway. Anchor it to a period the board recognizes, quarterly or annually, so progress reviews line up with the plan.

3. Check feasibility

Confirm the lift is possible by naming which operational drivers have to move, by how much, and at what cost in headcount or technology. If a five-point lift requires FCR to climb ten points on the voice channel, decide whether that is achievable with current tooling before the number reaches a board slide. Feasibility turns ambition into a plan.

4. Confirm the lift is necessary

Tie the lift to a financial outcome, retention or renewal rate, for example; a lift with no linked outcome is a vanity figure. If a two-point NPS increase does not translate into measurable retention, churn, or expansion revenue, you cannot defend the investment when priorities compete. The link is what turns a loyalty score into a business case.

5. Scale targets across business units

Set different targets for units with different baselines and customer bases, since a single company-wide number can hide the unit that is slipping. An enterprise segment starting at 40 and an SMB segment starting at 15 need separate lifts and separate feasibility checks. Rolling them into a single number lets the strong unit mask the weak one.

The sequence also includes a desirable ceiling, the point where the next NPS point costs more than it returns. Past that point, spending on the score is spending on a slide. Put the ceiling in the plan so the target is a band with a floor and a stop, not a single number.

6. Segment by handler type

Measure episode NPS separately for AI-only contacts, hybrid contacts where an AI agent started the call and a human agent finished it, and human-only contacts. Separate scores show whether the automated path produces promoters and whether detractors appear at the handover.

That makes escalation design the variable to govern: automation rate shows how many calls an AI agent took, and the hybrid episode score shows what happened to the callers it couldn't finish.

Use NPS benchmarks by industry as context and move the drivers instead

Industry benchmarks belong in the appendix, not on the target line. The score that decides whether your CX program is working is your own prior-period NPS for the same type, the same population, and now the same handler mix, paired with the operational metrics that produced it. A detractor rating measures the distance between what the caller needed and what the phone channel delivered, and closing that distance is an operational job.

Parloa builds AI agents for the phone channel that resolve calls end to end, hand over to human agents with full context when escalation is the right call, and expose every routing and intent decision for review. The platform spans Build, Optimize, and Observe, so teams can test escalation logic before callers encounter it and monitor handler-type episode NPS after launch. Compliance certifications include ISO 27001:2022, ISO 17442:2020, SOC 2 Type 1 & 2, PCI DSS, HIPAA, and DORA.

Book a demo to see how AI agents move the drivers behind your NPS.

FAQs about NPS benchmarks by industry

Can we compare our NPS to a competitor's published score?

Only when the NPS type, the respondent population, and the survey period all match, and in practice they rarely do. A competitor's published figure may be a self-fielded relationship score with no disclosed sample, which makes it incomparable to your post-call score.

Should AI-handled calls be measured with the same NPS survey as human calls?

Use the same episode NPS question and segment the results by handler type. Keep all of it out of the relationship score, which comes from the brand survey. A blended figure hides whether the automated path or the handover is producing detractors.

How many survey responses do I need to trust a change in NPS?

Calculate a confidence interval for the response count before treating a change as reliable. A small sample produces a wide confidence interval, so a score that moves several points between periods may reflect who happened to answer rather than any change in service. Pool responses across a longer window or a larger unit before treating a move as a signal.

Is NPS still the right primary metric for an enterprise CX team?

Critics contest NPS as a sole primary metric, and the criticism that it cannot explain its own movements is fair. It holds as one loyalty signal when paired with the operational drivers behind it, plus a transactional measure taken after the contact. When teams use it that way, a quarterly change in NPS has a candidate cause before anyone has to guess.